Rapport final - Analyse des incidents liés à la police aux États-Unis d'Amérique¶

Mohamedlamine Amimi mohamedlamine.amimi@supinfo.com Théo Bezin theo.bezin@supinfo.com Alexis Lecomte alexis.lecomte@supinfo.com Maxence Pawlowski maxence.pawlowski@supinfo.com

In [1]:
# Calculs mathématiques
import math
import numpy

# Chargement et prise en charge des données
import pandas

# Prise en charge des données géographique
import geopandas
import folium
from folium import plugins

# Création de statistiques et de graphiques
import scipy.stats
import matplotlib.pyplot as pyplot
from matplotlib.ticker import NullFormatter
import plotly.express as px

# Autres
import os
import re
import time
from datetime import datetime
from tabulate import tabulate
import webbrowser
from colorama import Fore, Style

%matplotlib notebook
pyplot.style.use("ggplot")
pyplot.rcParams["figure.figsize"] = [17, 7]
pyplot.rcParams["figure.max_open_warning"] = 30

Analyse de "WashingtonPostDatabase.csv"¶

In [2]:
data_wp = pandas.read_csv("../data/WahsingtonPostDatabase.csv", parse_dates=["date"], date_parser=pandas.to_datetime)
data_wp.dtypes
Out[2]:
id                                  int64
name                               object
date                       datetime64[ns]
manner_of_death                    object
armed                              object
age                               float64
gender                             object
race                               object
city                               object
state                              object
signs_of_mental_illness              bool
threat_level                       object
flee                               object
body_camera                          bool
longitude                         float64
latitude                          float64
is_geocoding_exact                   bool
dtype: object

Nous vérifions la qualité du dataset.

In [3]:
data_wp.isna().sum()
Out[3]:
id                            0
name                        353
date                          0
manner_of_death               0
armed                       209
age                         413
gender                        7
race                       1323
city                          0
state                         0
signs_of_mental_illness       0
threat_level                  0
flee                        653
body_camera                   0
longitude                   711
latitude                    711
is_geocoding_exact            0
dtype: int64

Nous nous débarrassons des données nulles.

In [4]:
data_wp["name"] = data_wp["name"].fillna("Unknown") 				# replace NaN  by unknown
data_wp["armed"] = data_wp["armed"].fillna("Unknown") 				# replace NaN  by unknown
data_wp["gender"] = data_wp["gender"].fillna("Unknown") 			# replace NaN  by unknown
data_wp["race"] = data_wp["race"].fillna("Unknown") 				# replace NaN  by unknown
data_wp["threat_level"] = data_wp["threat_level"].fillna("Unknown") # replace NaN  by unknown
data_wp["flee"] = data_wp["flee"].fillna("Unknown") 				# replace NaN  by unknown
data_wp["latitude"] = data_wp["latitude"].fillna("Unknown") 		# replace NaN  by unknown
data_wp["longitude"] = data_wp["longitude"].fillna(0) 				# replace NaN  by unknown
age_mean = data_wp["age"].mean()
data_wp["age"] = data_wp["age"].fillna(round(age_mean))  			# replace NaN age by mean

data_wp.head()
Out[4]:
id name date manner_of_death armed age gender race city state signs_of_mental_illness threat_level flee body_camera longitude latitude is_geocoding_exact
0 3 Tim Elliot 2015-01-02 shot gun 53.0 M A Shelton WA True attack Not fleeing False -123.122 47.247 True
1 4 Lewis Lee Lembke 2015-01-02 shot gun 47.0 M W Aloha OR False attack Not fleeing False -122.892 45.487 True
2 5 John Paul Quintero 2015-01-03 shot and Tasered unarmed 23.0 M H Wichita KS False other Not fleeing False -97.281 37.695 True
3 8 Matthew Hoffman 2015-01-04 shot toy weapon 32.0 M W San Francisco CA True attack Not fleeing False -122.422 37.763 True
4 9 Michael Rodriguez 2015-01-04 shot nail gun 39.0 M H Evans CO False attack Not fleeing False -104.692 40.384 True

Nous vérifions les cases vides de chaque colonne de données.

In [5]:
data_wp.isna().sum()
Out[5]:
id                         0
name                       0
date                       0
manner_of_death            0
armed                      0
age                        0
gender                     0
race                       0
city                       0
state                      0
signs_of_mental_illness    0
threat_level               0
flee                       0
body_camera                0
longitude                  0
latitude                   0
is_geocoding_exact         0
dtype: int64
In [6]:
nb_data = data_wp.shape[1] * data_wp.shape[0]
nb_data
Out[6]:
123624
In [7]:
data_wp.isna().sum().sum() / nb_data * 100
Out[7]:
0.0

Nous avons 0 % de données manquantes après le nettoyage des données.

In [8]:
data_wp.duplicated().sum()
Out[8]:
0

Aucune duplication dans le fichier.

Skew - Médiane - Mode - Moyenne¶

Le coefficient d'asymétrie (pour la loi normale skew=0)

In [9]:
(data_wp.skew(numeric_only=True) - 0).sort_values()
Out[9]:
is_geocoding_exact        -20.029178
id                         -0.015711
age                         0.758072
signs_of_mental_illness     1.376367
longitude                   1.561354
body_camera                 2.073920
dtype: float64
In [10]:
data_wp.age.median()  # the middle valuex c'est 36
Out[10]:
36.0
In [11]:
data_wp.age.mode()  # the most repeated value est 37
Out[11]:
0    37.0
Name: age, dtype: float64
In [12]:
data_wp.age.mean()  # the average value age 37 ans
Out[12]:
37.12898789878988
In [13]:
wp_age_fig, wp_age_ax = pyplot.subplots()
_ = data_wp.age.plot(ax=wp_age_ax, kind="box", vert=False)
In [14]:
data_wp.race.value_counts(normalize=True) * 100  # Transformation 41 % de W et 21% B attention au moins 18% ne sont pas connus
Out[14]:
W          41.570407
B          21.933443
Unknown    18.193069
H          14.961496
A           1.443894
N           1.251375
O           0.646315
Name: race, dtype: float64
In [15]:
wp_race_fig, wp_race_ax = pyplot.subplots()
_ = data_wp.race.value_counts().plot(ax=wp_race_ax, kind="bar")
In [16]:
data_wp.gender.value_counts(normalize=True) * 100  # Transformation % 95% sont des hommes
Out[16]:
M          95.393289
F           4.510451
Unknown     0.096260
Name: gender, dtype: float64
In [17]:
wp_age2_fig, wp_age2_ax = pyplot.subplots()
_ = data_wp.hist("age", ax=wp_age2_ax)
In [18]:
wp_gender_fig, wp_gender_ax = pyplot.subplots()
_ = data_wp.gender.value_counts().plot(ax=wp_gender_ax, kind="pie", autopct="%1.2f%%")
In [19]:
data_wp.flee.value_counts(normalize=True) * 100  # Transformation % plus de la moitié des arrestation se font sans tentative de fuite
Out[19]:
Not fleeing    58.182068
Car            16.240374
Foot           12.953795
Unknown         8.979648
Other           3.644114
Name: flee, dtype: float64
In [20]:
wp_flee_fig, wp_flee_ax = pyplot.subplots()
_ = data_wp.flee.value_counts().plot(ax=wp_flee_ax, kind="pie", autopct="%1.2f%%")
In [21]:
data_wp.manner_of_death.value_counts(normalize=True) * 100  # Transformation %
Out[21]:
shot                95.338284
shot and Tasered     4.661716
Name: manner_of_death, dtype: float64
In [22]:
data_wp.armed.value_counts(normalize=True) * 100  # Transformation % # 57 % des personnes étaient armées d'un pistolet
Out[22]:
gun                        57.109461
knife                      14.425193
unarmed                     6.091859
vehicle                     3.616612
undetermined                3.492849
                             ...    
baseball bat and bottle     0.013751
pen                         0.013751
chainsaw                    0.013751
gun and sword               0.013751
machete and hammer          0.013751
Name: armed, Length: 100, dtype: float64
In [23]:
wp_armed_fig, wp_armed_ax = pyplot.subplots()
_ = data_wp.armed.value_counts().plot(ax=wp_armed_ax, kind="barh", figsize=(20, 12))
In [24]:
data_wp.state.value_counts(normalize=True) * 100  # Transformation % 14% des crimes se passes en CA suivit de 8 au TX et 6 a FL
Out[24]:
CA    14.576458
TX     8.800880
FL     6.339384
AZ     4.579208
GA     3.712871
CO     3.589109
OK     2.805281
NC     2.805281
TN     2.777778
OH     2.722772
WA     2.722772
MO     2.626513
NM     2.117712
LA     2.062706
PA     2.062706
IL     2.007701
AL     1.897690
IN     1.883938
NY     1.828933
KY     1.828933
VA     1.746425
MI     1.718922
NV     1.663916
OR     1.608911
SC     1.567657
WI     1.526403
AR     1.402640
MD     1.402640
MS     1.292629
UT     1.168867
NJ     1.168867
MN     1.100110
KS     0.907591
WV     0.907591
ID     0.838834
MA     0.728823
AK     0.673817
MT     0.632563
IA     0.618812
NE     0.508801
HI     0.508801
ME     0.412541
SD     0.330033
CT     0.302530
DC     0.288779
WY     0.288779
NH     0.275028
DE     0.233773
ND     0.206271
VT     0.151265
RI     0.068757
Name: state, dtype: float64
In [25]:
wp_state_plt = data_wp.state.value_counts()
wp_state_fig, wp_state_ax = pyplot.subplots()

_ = wp_state_plt.plot(ax=wp_state_ax, kind="barh", figsize=(20, 12)) # l'état comprenant le plus de crime est la Californie
In [26]:
state_data = data_wp.state.value_counts().rename_axis("state_code").reset_index(name="number_of_deaths")
state_data
Out[26]:
state_code number_of_deaths
0 CA 1060
1 TX 640
2 FL 461
3 AZ 333
4 GA 270
5 CO 261
6 OK 204
7 NC 204
8 TN 202
9 OH 198
10 WA 198
11 MO 191
12 NM 154
13 LA 150
14 PA 150
15 IL 146
16 AL 138
17 IN 137
18 NY 133
19 KY 133
20 VA 127
21 MI 125
22 NV 121
23 OR 117
24 SC 114
25 WI 111
26 AR 102
27 MD 102
28 MS 94
29 UT 85
30 NJ 85
31 MN 80
32 KS 66
33 WV 66
34 ID 61
35 MA 53
36 AK 49
37 MT 46
38 IA 45
39 NE 37
40 HI 37
41 ME 30
42 SD 24
43 CT 22
44 DC 21
45 WY 21
46 NH 20
47 DE 17
48 ND 15
49 VT 11
50 RI 5
In [27]:
data_wp["year"] = pandas.DatetimeIndex(data_wp["date"]).year  # regrouper la date par an pour avoir une vision plus globale
year_data = data_wp["year"].value_counts().rename_axis("year").reset_index(name="deaths").sort_values(by="deaths")
year_data
Out[27]:
year deaths
7 2022 281
6 2016 958
5 2017 981
4 2018 985
3 2015 994
2 2019 999
1 2020 1020
0 2021 1054
In [28]:
state_plt = data_wp.year.value_counts()
wp_state2_fig, wp_state2_ax = pyplot.subplots()

_ = state_plt.plot(ax=wp_state2_ax, kind="bar")
In [29]:
data_wp.head()
Out[29]:
id name date manner_of_death armed age gender race city state signs_of_mental_illness threat_level flee body_camera longitude latitude is_geocoding_exact year
0 3 Tim Elliot 2015-01-02 shot gun 53.0 M A Shelton WA True attack Not fleeing False -123.122 47.247 True 2015
1 4 Lewis Lee Lembke 2015-01-02 shot gun 47.0 M W Aloha OR False attack Not fleeing False -122.892 45.487 True 2015
2 5 John Paul Quintero 2015-01-03 shot and Tasered unarmed 23.0 M H Wichita KS False other Not fleeing False -97.281 37.695 True 2015
3 8 Matthew Hoffman 2015-01-04 shot toy weapon 32.0 M W San Francisco CA True attack Not fleeing False -122.422 37.763 True 2015
4 9 Michael Rodriguez 2015-01-04 shot nail gun 39.0 M H Evans CO False attack Not fleeing False -104.692 40.384 True 2015

Analyse de "PoliceDeaths.csv"¶

In [30]:
data_pd = pandas.read_csv("../data/PoliceDeaths.csv", index_col=None, parse_dates=["date"], date_parser=pandas.to_datetime)
data_pd["state"] = data_pd["state"].map(lambda state: state.strip())

pd_state_data = pandas.DataFrame(
	data=data_pd["state"].value_counts().reset_index().values,
	columns=["State", "Number of deaths"]
).sort_values("State", ascending=True)

data_pd
Out[30]:
person dept eow cause cause_short date year canine dept_name state
0 Constable Darius Quimby Albany County Constable's Office, NY EOW: Monday, January 3, 1791 Cause of Death: Gunfire Gunfire 1791-01-03 1791 False Albany County Constable's Office NY
1 Sheriff Cornelius Hogeboom Columbia County Sheriff's Office, NY EOW: Saturday, October 22, 1791 Cause of Death: Gunfire Gunfire 1791-10-22 1791 False Columbia County Sheriff's Office NY
2 Deputy Sheriff Isaac Smith Westchester County Sheriff's Department, NY EOW: Thursday, May 17, 1792 Cause of Death: Gunfire Gunfire 1792-05-17 1792 False Westchester County Sheriff's Department NY
3 Marshal Robert Forsyth United States Department of Justice - United S... EOW: Saturday, January 11, 1794 Cause of Death: Gunfire Gunfire 1794-01-11 1794 False United States Department of Justice - United S... US
4 Sheriff Robert Maxwell Greenville County Sheriff's Office, SC EOW: Sunday, November 12, 1797 Cause of Death: Gunfire Gunfire 1797-11-12 1797 False Greenville County Sheriff's Office SC
... ... ... ... ... ... ... ... ... ... ...
22795 K9 Bruno Amarillo Police Department, TX EOW: Sunday, June 12, 2016 Cause of Death: Accidental Accidental 2016-06-12 2016 True Amarillo Police Department TX
22796 K9 Lazer United States Department of Homeland Security ... EOW: Monday, June 20, 2016 Cause of Death: Heat exhaustion Heat exhaustion 2016-06-20 2016 True United States Department of Homeland Security ... US
22797 K9 Tyson Fountain County Sheriff's Office, IN EOW: Monday, June 27, 2016 Cause of Death: Heat exhaustion Heat exhaustion 2016-06-27 2016 True Fountain County Sheriff's Office IN
22798 K9 Credo Long Beach Police Department, CA EOW: Tuesday, June 28, 2016 Cause of Death: Gunfire (Accidental) Gunfire (Accidental) 2016-06-28 2016 True Long Beach Police Department CA
22799 K9 Roscoe Emmett Police Department, ID EOW: Friday, July 1, 2016 Cause of Death: Automobile accident Automobile accident 2016-07-01 2016 True Emmett Police Department ID

22800 rows × 10 columns

In [31]:
data_pd.head()
Out[31]:
person dept eow cause cause_short date year canine dept_name state
0 Constable Darius Quimby Albany County Constable's Office, NY EOW: Monday, January 3, 1791 Cause of Death: Gunfire Gunfire 1791-01-03 1791 False Albany County Constable's Office NY
1 Sheriff Cornelius Hogeboom Columbia County Sheriff's Office, NY EOW: Saturday, October 22, 1791 Cause of Death: Gunfire Gunfire 1791-10-22 1791 False Columbia County Sheriff's Office NY
2 Deputy Sheriff Isaac Smith Westchester County Sheriff's Department, NY EOW: Thursday, May 17, 1792 Cause of Death: Gunfire Gunfire 1792-05-17 1792 False Westchester County Sheriff's Department NY
3 Marshal Robert Forsyth United States Department of Justice - United S... EOW: Saturday, January 11, 1794 Cause of Death: Gunfire Gunfire 1794-01-11 1794 False United States Department of Justice - United S... US
4 Sheriff Robert Maxwell Greenville County Sheriff's Office, SC EOW: Sunday, November 12, 1797 Cause of Death: Gunfire Gunfire 1797-11-12 1797 False Greenville County Sheriff's Office SC

Infos¶

In [32]:
data_pd.shape
Out[32]:
(22800, 10)
In [33]:
data_pd.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 22800 entries, 0 to 22799
Data columns (total 10 columns):
 #   Column       Non-Null Count  Dtype         
---  ------       --------------  -----         
 0   person       22800 non-null  object        
 1   dept         22800 non-null  object        
 2   eow          22800 non-null  object        
 3   cause        22800 non-null  object        
 4   cause_short  22800 non-null  object        
 5   date         22800 non-null  datetime64[ns]
 6   year         22800 non-null  int64         
 7   canine       22800 non-null  bool          
 8   dept_name    22800 non-null  object        
 9   state        22800 non-null  object        
dtypes: bool(1), datetime64[ns](1), int64(1), object(7)
memory usage: 1.6+ MB

22800 personnes en tout avec 9 informations par personne.

Données manquantes¶

In [34]:
data_pd.isna().sum()
Out[34]:
person         0
dept           0
eow            0
cause          0
cause_short    0
date           0
year           0
canine         0
dept_name      0
state          0
dtype: int64

Il ne nous manque aucune donnée.

Nombre de données totales¶

In [35]:
nb_data = data_pd.shape[1] * data_pd.shape[0]
nb_data
Out[35]:
228000

Données dupliquées¶

In [36]:
data_pd.duplicated().sum()
Out[36]:
0

Nous n'avons aucunes données dupliquées

Moyenne, écart-type, minimum, etc¶

In [37]:
data_pd.describe()
Out[37]:
year
count 22800.000000
mean 1951.751272
std 38.233234
min 1791.000000
25% 1924.000000
50% 1952.000000
75% 1983.000000
max 2016.000000

Nombre de valeurs canine¶

In [38]:
data_pd.canine.value_counts()
Out[38]:
False    22537
True       263
Name: canine, dtype: int64
In [39]:
data_pd.canine.value_counts(normalize=True)*100 #Transformation %
Out[39]:
False    98.846491
True      1.153509
Name: canine, dtype: float64
In [40]:
pd_canine_fig, pd_canine_ax = pyplot.subplots()
_ = data_pd.canine.value_counts().plot(ax=pd_canine_ax, kind="pie", autopct="%1.2f%%")

Répartition par État¶

In [41]:
pd_state_fig, pd_state_ax = pyplot.subplots()
_ = data_pd.state.value_counts().plot(ax=pd_state_ax, kind="pie")

L'état comprenant le plus de morts de policiers est le Texas, suivi par la Californie et l'état de New York.

Répartition par cause de décès¶

In [42]:
pd_cause_fig, pd_cause_ax = pyplot.subplots()
_ = data_pd.cause_short.value_counts().plot(ax=pd_cause_ax, kind="pie")

On peut voir que la cause de mort principale sont les armes à feu, ensuite les accidents de voiture.

Impact de la localisation (par état et par emplacement)¶

In [43]:
data_pd["state"].value_counts()
Out[43]:
TX    1880
CA    1639
NY    1574
US    1225
IL    1073
PA     901
OH     834
FL     820
KY     772
GA     698
MO     666
TN     607
MI     574
NC     549
AL     537
VA     523
OK     496
NJ     485
LA     470
IN     434
PR     363
SC     362
MA     360
MD     325
CO     305
WA     290
AR     284
RR     280
WI     275
AZ     273
MS     261
KS     257
MN     244
WV     208
IA     197
OR     182
NM     151
CT     144
DC     128
UT     128
MT     128
NV     120
NE     110
ME      86
ID      64
WY      57
HI      56
ND      55
SD      53
TR      51
RI      50
AK      48
NH      46
DE      41
VT      26
VI      14
GU      12
CZ       3
AS       3
MP       3
Name: state, dtype: int64
In [44]:
pd_state_data
Out[44]:
State Number of deaths
51 AK 48
14 AL 537
26 AR 284
58 AS 3
29 AZ 273
1 CA 1639
24 CO 305
37 CT 144
57 CZ 3
38 DC 128
53 DE 41
7 FL 820
9 GA 698
56 GU 12
46 HI 56
34 IA 197
44 ID 64
4 IL 1073
19 IN 434
31 KS 257
8 KY 772
18 LA 470
22 MA 360
23 MD 325
43 ME 86
12 MI 574
32 MN 244
10 MO 666
59 MP 3
30 MS 261
40 MT 128
13 NC 549
47 ND 55
42 NE 110
52 NH 46
17 NJ 485
36 NM 151
41 NV 120
2 NY 1574
6 OH 834
16 OK 496
35 OR 182
5 PA 901
20 PR 363
50 RI 50
27 RR 280
21 SC 362
48 SD 53
11 TN 607
49 TR 51
0 TX 1880
3 US 1225
39 UT 128
15 VA 523
55 VI 14
54 VT 26
25 WA 290
28 WI 275
33 WV 208
45 WY 57
In [45]:
pd_state2_fig = px.choropleth(
	pd_state_data,
    locations="State",
    color="Number of deaths", color_continuous_scale="spectral_r",
    locationmode="USA-states",
    labels={"Nombre de morts": "number_of_deaths"},
    scope="usa",
	title="Nombre de morts totales de policiers par état aux États-Unis"
)

pd_state2_fig.show()

On observe sur cette carte la répartition géographique du nombre de morts de policiers selon l'état. On ne peut pas vraiment conclure d'une corrélation entre la situation géographique d'un état et son nombre de policiers tués en service.

In [46]:
pd_state3_fig = px.bar(pd_state_data, x="State", y="Number of deaths", text_auto=".2s", title="Nombre de morts totales de policiers par état aux États-Unis")
pd_state3_fig.update_traces(textfont_size=20, textangle=0, textposition="outside", cliponaxis=False)
pd_state3_fig.show()

L'état du Texas est l'état qui comporte le plus de morts de policiers totales au cours de ce siècle dernier avec plus de 1900 décès totaux.

Étude par an¶

In [47]:
data_pd["year"].value_counts()
Out[47]:
1930    323
1927    305
1932    301
1931    296
1924    289
       ... 
1824      1
1827      1
1828      1
1830      1
1792      1
Name: year, Length: 202, dtype: int64
In [48]:
year_data = data_pd["year"].value_counts().rename_axis("year").reset_index(name="number_of_deaths").sort_values(by="year")
year_data
Out[48]:
year number_of_deaths
181 1791 2
201 1792 1
186 1794 1
187 1797 1
188 1804 1
... ... ...
67 2012 157
85 2013 141
62 2014 165
68 2015 157
121 2016 76

202 rows × 2 columns

In [49]:
pd_deaths_fig = px.bar(year_data, x="year", y="number_of_deaths", text_auto=".2s", title="Nombre de morts totales de policiers au cours du siècle dernier")
pd_deaths_fig.update_traces(textfont_size=20, textangle=0, textposition="outside", cliponaxis=False)
pd_deaths_fig.show()

On remarque que l'année 1930 a été la plus meurtrière pour les policiers aux États-Unis avec 323 morts de policiers pendant leur service. Cela peut s'expliquer par la surcharge de travail de la police d'État à cette époque. D'après un rapport de la police de New Jersey, "Dans les années 1930, il n'était pas rare que les policiers travaillent de douze à seize heures par jour, sans jour de repos, parfois plusieurs semaines d'affilée, pour accomplir leur mission de protection et de service."

https://nj.gov/njsp/about/history/1930s.shtml

In [50]:
statebyyear_data = (data_pd["state"].where(data_pd["year"] == 1930)).value_counts()
statebyyear_data
Out[50]:
NY    37
IL    27
US    21
TX    18
OK    17
MI    15
PA    11
CA    11
NJ    10
TN     9
RR     9
VA     9
MA     8
MO     8
KY     7
OH     7
FL     7
IA     7
IN     6
NC     6
GA     6
KS     6
WI     6
MN     5
LA     4
WA     4
AL     4
WV     3
ME     3
OR     3
PR     3
AR     3
NE     3
CT     3
CO     2
RI     2
NM     2
DC     2
MS     2
ND     2
MD     1
AZ     1
HI     1
SC     1
MT     1
Name: state, dtype: int64

On peut observer que cette année-là, en 1930, c'est l'état de New York qui présente le plus de morts de policiers.

In [51]:
y_causes = pandas.DataFrame()
ny_causes = data_pd["cause_short"][(data_pd.year == 1930) & (data_pd.state.str.contains("NY"))].value_counts()
ny_causes= ny_causes.rename_axis("cause").reset_index(name="number_of_deaths")
ny_causes
Out[51]:
cause number_of_deaths
0 Gunfire 11
1 Motorcycle accident 9
2 Automobile accident 5
3 Struck by vehicle 4
4 Heart attack 2
5 Gunfire (Accidental) 1
6 Stabbed 1
7 Vehicular assault 1
8 Fall 1
9 Electrocuted 1
10 Vehicle pursuit 1
In [52]:
pd_cause2_fig = px.bar(ny_causes, x="cause", y="number_of_deaths", text_auto=".2s", title="Causes de mort de policier dans l'état de New York en 1930")
pd_cause2_fig.update_traces(textfont_size=12, textangle=0, textposition="outside", cliponaxis=False)
pd_cause2_fig.show()

La plupart de ces morts sont dues à des armes à feu, ce sont donc des policiers tombés lors de leur service durant des fusillades ou attaques à armes à feu.

Etude des modalités des incidents¶

In [53]:
causes_data = data_pd["cause_short"].value_counts()
causes_data = causes_data.rename_axis("cause").reset_index(name="number_of_deaths")
causes_data
Out[53]:
cause number_of_deaths
0 Gunfire 12067
1 Automobile accident 2348
2 Motorcycle accident 1134
3 Heart attack 977
4 Vehicular assault 888
5 Struck by vehicle 868
6 Vehicle pursuit 627
7 Assault 613
8 Gunfire (Accidental) 604
9 Stabbed 465
10 Aircraft accident 381
11 Drowned 262
12 Struck by train 254
13 Fall 197
14 Duty related illness 169
15 9/11 related illness 110
16 Terrorist attack 85
17 Electrocuted 83
18 Animal related 82
19 Accidental 80
20 Struck by streetcar 68
21 Fire 63
22 Heat exhaustion 61
23 Explosion 58
24 Bomb 57
25 Exposure to toxins 44
26 Weather/Natural disaster 33
27 Training accident 28
28 Structure collapse 23
29 Train accident 18
30 Exposure 14
31 Boating accident 13
32 Bicycle accident 13
33 Unidentified 7
34 Poisoned 4
35 Asphyxiation 2
In [54]:
pd_cause3_fig = px.bar(causes_data, x="cause", y="number_of_deaths", text_auto=".2s", title="Causes de mort de policier pendant leur service")
pd_cause3_fig.update_traces(textfont_size=12, textangle=0, textposition="outside", cliponaxis=False)
pd_cause3_fig.show()

Analyse de "FatalEncounters.csv"¶

Chargement des données¶

Nous importons le fichier CSV en spécifiant les types de chaque colonne. Ainsi, le chargement est plus rapide, car pandas n'as pas besoin de deviner le type de chacune des colonnes. Et cela permet aussi définir le type que nous souhaitons avoir, laissant ainsi moins de place au hasard.

In [55]:
data_fe = pandas.read_csv("../data/FatalEncounters.csv", dtype={
	"Unique ID": float,
	"Name": str,
	"Age": str,
	"Gender": "category",
	"Race": "category",
	"Race with imputations": "category",
	"URL of image (PLS NO HOTLINKS)": str,
	"Location of injury (address)": str,
	"Location of death (city)": str,
	"State": str,
	"Location of death (zip code)": str,
	"Location of death (country)": str,
	"Full Address": str,
	"Latitude": str,
	"Longitude": float,
	"Agency or agencies involved": str,
	"Highest level of force": "category",
	"UID Temporary": str,
	"Name Temporary": str,
	"Armed/Unarmed": str,
	"Alleged weapon": str,
	"Aggressive physical movement": str,
	"Fleeing/Not fleeing": str,
	"Description Temp": str,
	"URL Temp": str,
	"Brief description": str,
	"Dispositions/Exclusions INTERNAL USE, NOT FOR ANALYSIS": object,
	"Intended use of force (Developing)": str,
	"Supporting document link": str,
	"Foreknowledge of mental illness? INTERNAL USE, NOT FOR ANALYSIS": object,
	"Unnamed: 32": float,
	"Unnamed: 33": float,
	"Unique ID formula": float,
	"Unique identifier (redundant)": float
}, parse_dates=[" Date of injury resulting in death (month/day/year)"], date_parser=pandas.to_datetime)

Nous pouvons noter que certaines des colonnes n'ont pas un type qui semble adéquat. Par exemple, la latitude qui est une str plutôt qu'un float. En effet, cette colonne comporte des données erronées qui empêche d'interpréter chacune de ses valeurs comme un float. Nous réglerons le problème après avoir préparé les données.

Premier aperçu¶

In [56]:
print(f"Shape : {Fore.LIGHTGREEN_EX}{data_fe.shape}{Fore.RESET}")
print(data_fe.head(n=5))
print(f"Type des colonnes : {Style.DIM}{Fore.WHITE}\n{data_fe.dtypes}{Style.RESET_ALL}")

Shape : (31498, 36)
   Unique ID                   Name Age  Gender                     Race  \
0    25746.0        Samuel H. Knapp  17    Male  European-American/White   
1    25747.0         Mark A. Horton  21    Male   African-American/Black   
2    25748.0  Phillip A. Blurbridge  19    Male   African-American/Black   
3    25749.0             Mark Ortiz  23    Male          Hispanic/Latino   
4        1.0  LaTanya Janelle McCoy  24  Female   African-American/Black   

     Race with imputations Imputation probability  \
0  European-American/White            Not imputed   
1   African-American/Black            Not imputed   
2   African-American/Black            Not imputed   
3          Hispanic/Latino            Not imputed   
4   African-American/Black            Not imputed   

  URL of image (PLS NO HOTLINKS)  \
0                            NaN   
1                            NaN   
2                            NaN   
3                            NaN   
4                            NaN   

   Date of injury resulting in death (month/day/year)  \
0                                         2000-01-01    
1                                         2000-01-01    
2                                         2000-01-01    
3                                         2000-01-01    
4                                         2000-01-02    

  Location of injury (address)  ... URL Temp  \
0           27898-27804 US-101  ...      NaN   
1              Davison Freeway  ...      NaN   
2              Davison Freeway  ...      NaN   
3              600 W Cherry Ln  ...      NaN   
4         5700 block Mack Road  ...      NaN   

                                   Brief description  \
0  Samuel Knapp was allegedly driving a stolen ve...   
1  Two Detroit men killed when their car crashed ...   
2  Two Detroit men killed when their car crashed ...   
3  A motorcycle was allegedly being driven errati...   
4  LaTanya Janelle McCoy's car was struck from be...   

  Dispositions/Exclusions INTERNAL USE, NOT FOR ANALYSIS  \
0                                         Unreported       
1                                         Unreported       
2                                         Unreported       
3                                         Unreported       
4                                            Unknown       

  Intended use of force (Developing)  \
0                            Pursuit   
1                            Pursuit   
2                            Pursuit   
3                            Pursuit   
4                            Pursuit   

                            Supporting document link  \
0  https://drive.google.com/file/d/10DisrV8K5ReP1...   
1  https://drive.google.com/file/d/1-nK-RohgiM-tZ...   
2  https://drive.google.com/file/d/1-nK-RohgiM-tZ...   
3  https://drive.google.com/file/d/1qAEefRjX_aTtC...   
4  http://www.recordnet.com/article/20000110/A_NE...   

  Foreknowledge of mental illness? INTERNAL USE, NOT FOR ANALYSIS  \
0                                                 No                
1                                                 No                
2                                                 No                
3                                                 No                
4                                                 No                

   Unnamed: 32 Unnamed: 33 Unique ID formula Unique identifier (redundant)  
0          NaN         NaN               NaN                       25746.0  
1          NaN         NaN               NaN                       25747.0  
2          NaN         NaN               NaN                       25748.0  
3          NaN         NaN               NaN                       25749.0  
4          NaN         NaN               NaN                           1.0  

[5 rows x 36 columns]
Type des colonnes : 
Unique ID                                                                 float64
Name                                                                       object
Age                                                                        object
Gender                                                                   category
Race                                                                     category
Race with imputations                                                    category
Imputation probability                                                     object
URL of image (PLS NO HOTLINKS)                                             object
 Date of injury resulting in death (month/day/year)                datetime64[ns]
Location of injury (address)                                               object
Location of death (city)                                                   object
State                                                                      object
Location of death (zip code)                                               object
Location of death (county)                                                 object
Full Address                                                               object
Latitude                                                                   object
Longitude                                                                 float64
Agency or agencies involved                                                object
Highest level of force                                                   category
UID Temporary                                                              object
Name Temporary                                                             object
Armed/Unarmed                                                              object
Alleged weapon                                                             object
Aggressive physical movement                                               object
Fleeing/Not fleeing                                                        object
Description Temp                                                           object
URL Temp                                                                   object
Brief description                                                          object
Dispositions/Exclusions INTERNAL USE, NOT FOR ANALYSIS                     object
Intended use of force (Developing)                                         object
Supporting document link                                                   object
Foreknowledge of mental illness? INTERNAL USE, NOT FOR ANALYSIS            object
Unnamed: 32                                                               float64
Unnamed: 33                                                               float64
Unique ID formula                                                         float64
Unique identifier (redundant)                                             float64
dtype: object

Unicité des colonnes "Unique ID" et "Name"¶

Nous cherchons ici à vérifier que ces deux colonnes sont uniques. Nous nous attendons à ce que la colonne "Unique ID" soit unique. Cependant, le nom peut comporter quelques doublons.

In [57]:
def is_column_unique(col_name, display_name=None):
	display_name = display_name if display_name else col_name
	is_unique = data_fe[col_name].is_unique
	print(f"Est-ce que \"{display_name}\" est unique ? : {Fore.LIGHTGREEN_EX}{is_unique}{Fore.RESET}")

	if not is_unique:
		columns_to_print = ["Unique ID"]

		if col_name not in columns_to_print:
			columns_to_print.append(col_name)

		duplicated = data_fe[data_fe.duplicated(subset=[col_name], keep=False)][columns_to_print]
		duplicated["Count"] = duplicated.groupby(col_name)["Unique ID"].transform("count")

		print((
			f"Les valeurs non uniques de \"{display_name}\" sont:\n"
			f"{Style.DIM}{Fore.WHITE}{duplicated.drop_duplicates([col_name])}{Style.RESET_ALL}"
		))

is_column_unique("Unique ID", display_name="id")
is_column_unique("Name")

Est-ce que "id" est unique ? : True
Est-ce que "Name" est unique ? : False
Les valeurs non uniques de "Name" sont:
       Unique ID                      Name  Count
22       20792.0   Name withheld by police   1391
98          59.0              Donald Moore      2
195      28746.0                John Doe 1      5
196      28747.0                John Doe 2      5
197      28748.0                John Doe 3      4
...          ...                       ...    ...
28227    28093.0       William Johnson Jr.      2
28248    28113.0              Dion Johnson      2
30806    30781.0              Eric Padilla      2
30820    30790.0     Devonte Dawayne Brown      2
31134    31134.0  Christopher Ryan Vasquez      2

[208 rows x 3 columns]

Valeurs manquantes par colonnes¶

Nous allons maintenant essayer de voir la quantité (en pourcentage) de valeurs manquantes pour chacune des colonnes de notre jeu de données.

In [58]:
print((
	"Pourcentage total de valeur manquante : "
	f"{Fore.LIGHTGREEN_EX}{round(((data_fe.isna().sum().sum() / (data_fe.shape[0] * data_fe.shape[1])) * 100), 2)}%{Fore.RESET}"
))

percent_missing = round((data_fe.isna().sum() * 100) / len(data_fe), 2)
df_missing = pandas.DataFrame({"Column name": data_fe.columns, "Missing": percent_missing.values})
df_missing.sort_values("Missing", ascending=False, inplace=True)

print((
	f"Pourcentage de valeur manquante par colonne :\n"
	f"{Style.DIM}{Fore.WHITE}{tabulate(df_missing, showindex=False, headers=df_missing.columns)}{Style.RESET_ALL}"
))

Pourcentage total de valeur manquante : 24.75%
Pourcentage de valeur manquante par colonne :
Column name                                                        Missing
---------------------------------------------------------------  ---------
Unnamed: 33                                                         100
Unnamed: 32                                                         100
Unique ID formula                                                    99.99
URL Temp                                                             89.79
Description Temp                                                     87.09
UID Temporary                                                        82.45
Name Temporary                                                       82.45
URL of image (PLS NO HOTLINKS)                                       53.24
Armed/Unarmed                                                        45.78
Fleeing/Not fleeing                                                  45.78
Alleged weapon                                                       45.78
Aggressive physical movement                                         45.77
Age                                                                   3.88
Imputation probability                                                2.82
Race with imputations                                                 2.77
Location of injury (address)                                          1.77
Location of death (zip code)                                          0.58
Gender                                                                0.46
Agency or agencies involved                                           0.25
Foreknowledge of mental illness? INTERNAL USE, NOT FOR ANALYSIS       0.2
Location of death (city)                                              0.11
Location of death (county)                                            0.05
Dispositions/Exclusions INTERNAL USE, NOT FOR ANALYSIS                0.01
Intended use of force (Developing)                                    0.01
Supporting document link                                              0.01
Brief description                                                     0.01
Highest level of force                                                0.01
Unique ID                                                             0
Name                                                                  0
Longitude                                                             0
Latitude                                                              0
Full Address                                                          0
State                                                                 0
Date of injury resulting in death (month/day/year)                    0
Race                                                                  0
Unique identifier (redundant)                                         0

Nous allons ensuite supprimer toutes les colonnes avec plus de x% (threshold) de données manquantes.

In [59]:
# Drop all columns with more than `drop_threshold`% of missing data
drop_threshold = 50
drop_columns = df_missing[df_missing["Missing"] >= drop_threshold]["Column name"].tolist()

if len(drop_columns) > 0:
	print((
		f"{Fore.YELLOW}Suppression des colonnes avec plus de {drop_threshold}% de données manquantes. Suppression de "
		f"{drop_columns}...{Fore.RESET}"
	))
	drop_prev_shape = data_fe.shape
	data_fe.drop(columns=drop_columns, axis=1, inplace=True)
	drop_new_shape = data_fe.shape
	drop_count = drop_prev_shape[1] - drop_new_shape[1]

	drop_plural = "s" if drop_count > 1 else ""
	print(f"{Fore.YELLOW}{drop_count} colonne{drop_plural} supprimée{drop_plural}. Nouvelle \"shape\" : {drop_new_shape}{Fore.RESET}")

Suppression des colonnes avec plus de 50% de données manquantes. Suppression de ['Unnamed: 33', 'Unnamed: 32', 'Unique ID formula', 'URL Temp', 'Description Temp', 'UID Temporary', 'Name Temporary', 'URL of image (PLS NO HOTLINKS)']...
8 colonnes supprimées. Nouvelle "shape" : (31498, 28)

Valeurs dupliquées¶

Pour clore ce premier aperçu, nous allons compter le nombre de données dupliquées.

In [60]:
duplicated_data = data_fe.duplicated().sum()
print((
	f"Données dupliquées : {Fore.LIGHTGREEN_EX}{duplicated_data}{Fore.RESET}"
	f" ({round((duplicated_data * 100) / data_fe.shape[0], 1)}%)"
))

Données dupliquées : 0 (0.0%)

Préparation des données¶

Avant toute analyse approfondie des données, il faut les préparer. Et notamment régler les problèmes de valeur pour certaines colonnes.

Suppression des colonnes à ne pas utiliser¶

Certaines colonnes comportent dans leur nom la mention "INTERNAL USE, NOT FOR ANALYSIS". Pour respecter cette mention, nous supprimons ces colonnes.

In [61]:
data_fe.drop(list(data_fe.filter(regex="INTERNAL USE, NOT FOR ANALYSIS")), axis=1, inplace=True)

Préparation de la colonne "Age"¶

Cette colonne ne pouvait pas être mise sous le type float car certaines de ses valeurs n'était pas numériques.

Certaines valeurs sont des intervalles d'âge. Deux choix sont alors possibles. Supprimer les lignes concernées ou transformer l'intervalle en moyenne. Nous avons choisi ici la deuxième solution. Pour régler le souci, nous appliquons une fonction de transformation sur la colonne.

In [62]:
def _transform_age(age):
	if isinstance(age, str):
		if "-" in age and not age.startswith("-"):
			age_range = age.split("-")
			return round(float((int(age_range[0]) + int(age_range[1])) / 2), 0)
		else:
			return round(float(age), 0)

	return round(age, 0)


data_fe["Age"] = data_fe["Age"].map(lambda age: _transform_age(age))

Préparation de la colonne "Latitude"¶

Certaines latitudes comportent des caractères invalides. La solution la plus simple les concernant et de supprimer les lignes en question.

In [63]:
wrong_latitude_index = []

def _transform_latitude(lat):
	if isinstance(lat, str):
		try:
			return float(lat)
		except ValueError:
			wrong_latitude_index.append(data_fe.index[data_fe["Latitude"] == lat])
			return lat

	return lat

data_fe["Latitude"] = data_fe["Latitude"].map(lambda lat: _transform_latitude(lat))
data_fe.drop([index for sublist in wrong_latitude_index for index in sublist], inplace=True)
data_fe = data_fe.astype({"Latitude": float})

Préparation de la colonne "Race"¶

Cette colonne pose deux problèmes. Le premier est que certaines cellules sont vides où contienne des valeurs incorrectes. Le second concerne la manière dont les ethnies sont renseignée. Chaque personne est caractérisée par une série d'ethnies séparées par un slash (/). Cependant, ce format rend compliqué l'analyse des données. Nous allons donc procéder à un One Hot Encoding afin de remplacer cette colonne par plusieurs colonnes correspondant à chacune des ethnies. Les valeurs de ces colonnes seront binaire, c'est-à-dire 0 où 1.

In [64]:
race_uncertain = "Race unspecified"
data_fe["Race"].fillna(race_uncertain, inplace=True)
data_fe["Race"] = data_fe["Race"].map(lambda race: race.title())
data_fe.loc[(data_fe["Race"] == "Christopher Anthony Alexander"), "Race"] = race_uncertain
data_fe = pandas.concat([
	data_fe.drop("Race", axis=1),
	data_fe["Race"].str.get_dummies(sep="/").add_prefix("Race_")
], axis=1)

Préparation de la colonne "Armed/Unarmed"¶

Cette colonne, tout comme "Fleeing/Not fleeing", et une colonne qui comporte un nombre important de valeurs manquantes. Cependant, elle peut être très utile à l'étude. Pour chacune des personnes, cette colonne à pour valeur "Armed" (armé), "Unarmed" (non armé) ou "Uncertain" (incertain). Nous allons donc remplacer les valeurs incorrectes ou nulles par "Uncertain".

In [65]:
armed_uncertain = "Uncertain"

def _transform_is_armed(is_armed):
	if isinstance(is_armed, str):
		if (is_armed.lower() == "armed") or (is_armed.lower() == "unarmed"):
			return is_armed
		elif re.match(r"duplicate|none", is_armed, flags=re.IGNORECASE):
			return armed_uncertain
		else:
			return armed_uncertain
	else:
		return armed_uncertain

data_fe["Armed/Unarmed"] = data_fe["Armed/Unarmed"].map(lambda is_armed: _transform_is_armed(is_armed))
data_fe = data_fe.astype({"Armed/Unarmed": "category"})

Préparation de la colonne "Fleeing/Not fleeing"¶

Nous avons le même intérêt pour cette colonne que "Armed/Unarmed", et elle comporte le même problème. Nous allons donc procéder de manière similaire.

In [66]:
fleeing_uncertain = "Uncertain"

def _transform_is_fleeing(is_fleeing):
	if isinstance(is_fleeing, str):
		if re.match(r"duplicate|none", is_fleeing, flags=re.IGNORECASE):
			return fleeing_uncertain
		else:
			parts = is_fleeing.lower().split("/")

			if "fleeing" in parts:
				return "Fleeing"
			elif "not fleeing" in parts:
				return "Not fleeing"
			elif fleeing_uncertain.lower() in parts:
				return fleeing_uncertain
			else:
				return fleeing_uncertain
	else:
		return fleeing_uncertain

data_fe["Fleeing/Not fleeing"] = data_fe["Fleeing/Not fleeing"].map(lambda is_fleeing: _transform_is_fleeing(is_fleeing))
data_fe = data_fe.astype({"Fleeing/Not fleeing": "category"})

Préparation de la colonne " Date of injury resulting in death (month/day/year)"¶

Cette colonne comporte un espace en trop au début. Nous réglons le problème et en profitons pour la renommer en quelque-chose de plus concis.

In [67]:
data_fe.rename({" Date of injury resulting in death (month/day/year)": "Date of death"}, axis=1, inplace=True)

Maintenant que nos transformations sont terminées, nous pouvons commencer à exploiter ce jeu de données pour une analyse statistique.

Analyse de la fréquence¶

In [68]:
freq_columns = ["Gender", "State", "Armed/Unarmed", "Fleeing/Not fleeing"]
freq_n_cols = 2
freq_n_rows = math.ceil(len(freq_columns) / freq_n_cols)
freq_fig = pyplot.figure()
freq_fig.suptitle("Fréquence des données qualitative")

for index, col in enumerate(freq_columns):
	ax = pyplot.subplot(freq_n_rows, freq_n_cols, index + 1)

	data_count = data_fe[[col]].value_counts()
	labels = [label[0] for label in data_count.index.tolist()]
	bars = data_count.plot(ax=ax, kind="bar", legend=True, label=col.title())
	ax.set_xlabel("")
	ax.set_xticks(ticks=numpy.arange(0, len(data_count)), labels=labels, rotation=45, fontsize=10)

	for bars in ax.containers:
		ax.bar_label(bars)

pyplot.tight_layout()
pyplot.show()

Analyse des champs numériques : Moyenne, Médiane, Mode, Écart-Type, Asymétrie (Skewness), Kurtosis¶

Notre jeu de données comporte assez peu de données numériques.

In [69]:
print(data_fe.dtypes)

Unique ID                                    float64
Name                                          object
Age                                          float64
Gender                                      category
Race with imputations                       category
Imputation probability                        object
Date of death                         datetime64[ns]
Location of injury (address)                  object
Location of death (city)                      object
State                                         object
Location of death (zip code)                  object
Location of death (county)                    object
Full Address                                  object
Latitude                                     float64
Longitude                                    float64
Agency or agencies involved                   object
Highest level of force                      category
Armed/Unarmed                               category
Alleged weapon                                object
Aggressive physical movement                  object
Fleeing/Not fleeing                         category
Brief description                             object
Intended use of force (Developing)            object
Supporting document link                      object
Unique identifier (redundant)                float64
Race_African-American                          int64
Race_Alaskan                                   int64
Race_Asian                                     int64
Race_Black                                     int64
Race_Black African-American                    int64
Race_Black Not Imputed                         int64
Race_European-American                         int64
Race_Hispanic                                  int64
Race_Latino                                    int64
Race_Middle Eastern                            int64
Race_Native American                           int64
Race_Pacific Islander                          int64
Race_Race Unspecified                          int64
Race_Race unspecified                          int64
Race_White                                     int64
dtype: object

Le "Unique ID" n'est qu'un identifiant et une analyse de moyenne sur une latitude et une longitude n'a pas vraiment de sens. La seule colonne analysée sera alors "Age".

In [70]:
numerical_columns = ["Age"]

for column in numerical_columns:
	# noinspection PyTypeChecker
	mmmssk_fig, (mmmssk_box_ax, mmmssk_hist_ax) = pyplot.subplots(
		nrows=2, ncols=1,
		sharex=True,
		gridspec_kw={"height_ratios": (0.2, 1)}
	)
	mmmssk_fig.suptitle(f"Statistiques de la colonne \"{column}\"")

	mean = data_fe[column].mean()
	median = data_fe[column].median()
	mode = data_fe[column].mode().values[0]

	data_fe[column].plot(ax=mmmssk_box_ax, kind="box", vert=False, legend=True)
	mmmssk_box_ax.axvline(mean, color="darkred", linestyle="--")
	mmmssk_box_ax.axvline(median, color="darkgreen", linestyle="-")
	mmmssk_box_ax.axvline(mode, color="darkblue", linestyle="-")

	data_fe[column].plot(ax=mmmssk_hist_ax, kind="hist", legend=True)
	mmmssk_hist_ax.axvline(mean, color="darkred", linestyle="--", label="Moyenne")
	mmmssk_hist_ax.axvline(median, color="darkgreen", linestyle="-", label="Médiane")
	mmmssk_hist_ax.axvline(mode, color="darkblue", linestyle="-", label="Mode")

	mmmssk_box_ax.set_yticklabels([""])
	mmmssk_hist_ax.legend()

	standard_deviation = data_fe[column].std()
	skewness = data_fe[column].skew()
	kurtosis = data_fe[column].kurtosis()

	pyplot.figtext(0.68, 0.58, f"Autres stats :", fontsize=10, fontweight="bold")
	pyplot.figtext(0.69, 0.54, f"Écart type : {round(standard_deviation, 3)}", fontsize=10)
	pyplot.figtext(0.69, 0.5, f"Skewness : {round(skewness, 3)}", fontsize=10)
	pyplot.figtext(0.69, 0.46, f"Kurtosis : {round(kurtosis, 3)}", fontsize=10)

	pyplot.tight_layout()
	pyplot.subplots_adjust(right=0.67)
	pyplot.show()

Répartition des confrontations fatales entre citoyen et policier¶

Nous allons chercher à voir la répartition de ces confrontations selon plusieurs variables. Nous pourrons ainsi voir quelles sont les personnes les plus touchés par ça.

In [71]:
repart_fig = pyplot.figure()
repart_size = (2, 2)
repart_grid = repart_fig.add_gridspec(repart_size[0], repart_size[1])

_ = repart_fig.suptitle("Répartition de rencontres fatales")

Répartition par âge et par sexe¶

In [72]:
male_labels = data_fe["Age"][data_fe["Gender"] == "Male"].drop_duplicates(keep="first", inplace=False).dropna().sort_values()
male_values = data_fe["Age"][data_fe["Gender"] == "Male"].value_counts().sort_index()

female_labels = data_fe["Age"][data_fe["Gender"] == "Female"].drop_duplicates(keep="first", inplace=False).dropna().sort_values()
female_values = data_fe["Age"][data_fe["Gender"] == "Female"].value_counts().sort_index()

trans_labels = data_fe["Age"][data_fe["Gender"] == "Transgender"].drop_duplicates(keep="first", inplace=False).dropna().sort_values()
trans_values = data_fe["Age"][data_fe["Gender"] == "Transgender"].value_counts().sort_index()

gender_ax = pyplot.subplot(repart_grid[0, :-1])
gender_ax.plot(male_labels, male_values.tolist(), label="Homme")
gender_ax.plot(female_labels, female_values.tolist(), label="Femme")
gender_ax.plot(trans_labels, trans_values.tolist(), label="Transgenre")

gender_ax.set_xlabel("Âge")
_ = gender_ax.legend()

Répartition par ethnies¶

In [73]:
# Revert one hot encoding
ethnic_data = data_fe[[column for column in data_fe if column.startswith("Race_")]].idxmax(1)
ethnic_data = ethnic_data.map(lambda ethnic: ethnic.replace("Race_", "").title())

ethnic_values = ethnic_data.value_counts().sort_values(ascending=False)
ethnic_labels = pandas.Series(data=ethnic_values.index, index=ethnic_values)

ethnic_ax = pyplot.subplot(repart_grid[1, :-1])
ethnic_ax.bar(ethnic_labels, ethnic_values.tolist())

_ = ethnic_ax.set_xticks(
	ticks=numpy.arange(0, len(ethnic_labels)),
	labels=ethnic_labels,
	rotation=45, fontsize=8, ha="right"
)

Répartition par État¶

In [74]:
# Fatal encounters per state
def autopct_func(pct):
	return f"{round(pct, 2)}%" if pct > 5 else ""

# Get states
states = data_fe["State"].sort_values()
states_values = states.value_counts().sort_values(ascending=False)

# Merge states with low values
states_threshold = 1
states_to_merge = states_values[states_values <= ((states_values.sum() * states_threshold) / 100)]
states_values = pandas.concat([
	states_values.drop(states_values.tail(n=len(states_to_merge)).index),
	pandas.Series(data=sum(states_to_merge), index=["Reste"])
])

# Get states labels
states_labels = pandas.Series(data=states_values.index, index=states_values)

# Plot
states_ax = pyplot.subplot(repart_grid[0:, -1])
states_ax.pie(states_values.tolist(), labels=states_labels, autopct=autopct_func, textprops={"fontsize": 8})
states_ax.axis("equal")
_ = states_ax.set_title("Répartition par État")

Enfin, il ne reste qu'à afficher les graphiques.

In [75]:
pyplot.tight_layout()
pyplot.show()

Test du khi-deux¶

In [76]:
# Contingency table
contingency = pandas.crosstab(
	index=data_fe["Fleeing/Not fleeing"],
	columns=data_fe["Armed/Unarmed"],
	normalize="index"
).round(4)
contingency_numpy = contingency.to_numpy()

chi_labels_armed = contingency.columns.tolist()
chi_labels_fleeing = contingency.index.tolist()

# Plot to heatmap
chi_fig, chi_ax = pyplot.subplots()

chi_ax.imshow(contingency_numpy)
chi_ax.set_xticks(numpy.arange(len(chi_labels_armed)), labels=chi_labels_armed)
chi_ax.set_yticks(numpy.arange(len(chi_labels_fleeing)), labels=chi_labels_fleeing)

for i in range(len(chi_labels_fleeing)):
	for j in range(len(chi_labels_armed)):
		chi_ax.text(j, i, contingency_numpy[i, j], ha="center", va="center", color="w")

# Interpret statistics
stat, pvalue, dof, expected = scipy.stats.chi2_contingency(contingency)

print(f"Degrés de liberté : {Fore.LIGHTGREEN_EX}{dof}{Fore.RESET}")
print(f"Fréquence attendue : {Fore.LIGHTGREEN_EX}{expected}{Fore.RESET}")

prob = 0.95
critical = scipy.stats.chi2.ppf(prob, dof)
print((
	f"Interprétation de test-statistic: "
	f"{Style.DIM}{Fore.WHITE}(Probability: {round(prob, 3)} | Critical: {round(critical, 3)} | Stat: {round(stat, 3)}) "
	f"{Style.RESET_ALL}{Fore.LIGHTGREEN_EX}>>> Dépendant: {'Oui' if (abs(stat) >= critical) else 'Non'}{Fore.RESET}"
))

alpha = 1.0 - prob
print((
	f"Interprétation de p-value: "
	f"{Style.DIM}{Fore.WHITE}(Significance: {round(alpha, 3)} | P-Value: {round(pvalue, 3)}) "
	f"{Style.RESET_ALL}{Fore.LIGHTGREEN_EX}>>> Dépendant: {'Oui' if (pvalue <= alpha) else 'Non'}{Fore.RESET}"
))

# Finish plot
chi_ax.set_title(
	"Test du khi-deux : Indépendance de \"Armed/Unarmed\" et \"Fleeing/Not fleeing\"",
	loc="center",
	wrap=True
)
chi_ax.grid(False)
chi_fig.tight_layout()
chi_fig.subplots_adjust(top=0.85)
pyplot.show()

Degrés de liberté : 4
Fréquence attendue : [[0.38477949 0.26884229 0.34637821]
 [0.38477949 0.26884229 0.34637821]
 [0.38474101 0.26881541 0.34634357]]
Interprétation de test-statistic: (Probability: 0.95 | Critical: 9.488 | Stat: 3.038) >>> Dépendant: Non
Interprétation de p-value: (Significance: 0.05 | P-Value: 0.552) >>> Dépendant: Non

Analyse de "TaxPolicyCenter.xls"¶

Mise en forme de départ¶

Le fichier Excel nécessite un traitement de départ pour pouvoir exploiter les données. Pour quelques-uns de nos calculs, nous utiliserons la feuille '2019' qui contient les données les plus récentes.

In [77]:
tax_data = pandas.read_excel(
    "../data/TaxPolicyCenter.xls",
    skiprows=5, skipfooter=4, # On ne prend pas en compte les lignes informatiques du header et du footer
    header=[0,1,2], # Les noms des colonnes sont sur trois lignes, nous traiterons donc ces lignes comme une seule ligne
    sheet_name=None
)
tax_sample = tax_data["2019"]

La boucle réalisera des traitements sur chaque feuille du fichier. D'abord, nous renommons les colonnes (puisque le header se compose de trois lignes) pour obtenir des noms cohérents. Nous récupérons également quelques données qui seront utilisées plus tard pour les graphiques et certains calculs.

In [78]:
tax_total_data = 0
total_income_list = []
police_budget = []
budget_years = []
print("===================Data INFO=====================")
tax_sample.info() #71 états/collectivités avec 11 colonnes de dépenses
for sheet in tax_data :
    income = 0
    df = tax_data[sheet]
    df.columns = ["_".join(index_str) for index_str in df.columns.to_flat_index()]
    df.rename(columns={
        "Unnamed: 0_level_0_Region and State_Unnamed: 0_level_2": "Region and State","Unnamed: 1_level_0_Region and State_Total": "Total",
        "Unnamed: 2_level_0_Inter-_governmental": "Inter-governmental",
        "Unnamed: 3_level_0_Inter-_Total": "Direct Total",
        "Elementary_and Secondary_Education": "Elementary and Secondary Education",
        "Elementary_Higher_Education": "Higher Education",
        "Elementary_Public_Welfare": "Public Welfare",
        "Elementary_Health and_Hospitals": "Health and Hospitals",
        "Elementary_Health and_Highways": "Highways",
        "Elementary_Health and_Police": "Police",
        "Elementary_All_Other": "All Other",
        "Personal_Income [1]_(Millions)": "Personal Income [1] (Millions)"
    }, inplace=True)
    print("===================Données manquantes pour " + sheet + " ?=====================")
    print(df.isna().sum()) # On vérifie si des valeurs sont manquantes
    df.dropna(inplace=True) # Suppression des lignes avec une valeur manquantes
    tax_total_data += df.shape[1]*df.shape[0] # On ajoute le total des données de la feuille dans une variable
    total_income_list.insert(0,(df.iloc[0]["Personal Income [1] (Millions)"])) # On ajoute
    police_budget.insert(0,(df.iloc[0]["Personal Income [1] (Millions)"])/100*(df.iloc[0]["Police"]*100))
    budget_years.insert(0, sheet)
tax_sample["All years income"] = pandas.Series(total_income_list)
tax_sample["Police budget evolution"] = pandas.Series(police_budget)

===================Data INFO=====================
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 71 entries, 0 to 70
Data columns (total 12 columns):
 #   Column                                                      Non-Null Count  Dtype  
---  ------                                                      --------------  -----  
 0   (Unnamed: 0_level_0, Region and State, Unnamed: 0_level_2)  60 non-null     object 
 1   (Unnamed: 1_level_0, Region and State, Total)               60 non-null     float64
 2   (Unnamed: 2_level_0, Inter-, governmental)                  60 non-null     float64
 3   (Unnamed: 3_level_0, Inter-, Total)                         60 non-null     float64
 4   (Elementary, and Secondary, Education)                      60 non-null     float64
 5   (Elementary, Higher, Education)                             60 non-null     float64
 6   (Elementary, Public, Welfare)                               60 non-null     float64
 7   (Elementary, Health and, Hospitals)                         60 non-null     float64
 8   (Elementary, Health and, Highways)                          60 non-null     float64
 9   (Elementary, Health and, Police)                            60 non-null     float64
 10  (Elementary, All, Other)                                    60 non-null     float64
 11  (Personal, Income [1], (Millions))                          60 non-null     float64
dtypes: float64(11), object(1)
memory usage: 6.8+ KB
===================Données manquantes pour 2019 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2018 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2017 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2016 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2015 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2014 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2013 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2012 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2011 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2010 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2009 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2008 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2007 ?=====================
Region and State                      11
Total                                 11
Inter-governmental                    11
Direct Total                          11
Elementary and Secondary Education    11
Higher Education                      11
Public Welfare                        11
Health and Hospitals                  11
Highways                              11
Police                                11
All Other                             11
Personal Income [1] (Millions)        11
dtype: int64
===================Données manquantes pour 2006 ?=====================
Region and State                      11
Total                                 12
Inter-governmental                    12
Direct Total                          12
Elementary and Secondary Education    12
Higher Education                      12
Public Welfare                        12
Health and Hospitals                  12
Highways                              12
Police                                12
All Other                             12
Personal Income [1] (Millions)        12
dtype: int64
===================Données manquantes pour 2005 ?=====================
Region and State                      11
Total                                 12
Inter-governmental                    12
Direct Total                          12
Elementary and Secondary Education    12
Higher Education                      12
Public Welfare                        12
Health and Hospitals                  12
Highways                              12
Police                                12
All Other                             12
Personal Income [1] (Millions)        12
dtype: int64
===================Données manquantes pour 2004 ?=====================
Region and State                      11
Total                                 12
Inter-governmental                    12
Direct Total                          12
Elementary and Secondary Education    12
Higher Education                      12
Public Welfare                        12
Health and Hospitals                  12
Highways                              12
Police                                12
All Other                             12
Personal Income [1] (Millions)        12
dtype: int64
In [79]:
print("===================Total des données=====================")
print(tax_total_data)

===================Total des données=====================
11520

Nous affichons des statistiques sur toutes les colonnes. Celles qui nous intéressent le plus sont les colonnes Police, All years income et Police budget evolution.

In [80]:
print("===================Description et stats=====================")
print(tax_sample.describe())

===================Description et stats=====================
           Total  Inter-governmental  Direct Total  \
count  60.000000           60.000000     60.000000   
mean   18.281756            0.007065     18.274691   
std     3.785900            0.020820      3.783119   
min     0.180093            0.000000      0.179921   
25%    16.517344            0.000000     16.517344   
50%    18.147631            0.000018     18.147577   
75%    19.794667            0.001935     19.794667   
max    28.095157            0.108886     28.095157   

       Elementary and Secondary Education  Higher Education  Public Welfare  \
count                           60.000000         60.000000       60.000000   
mean                             3.867793          1.767521        4.031424   
std                              0.797018          0.613697        1.352895   
min                              0.038711          0.016792        0.040109   
25%                              3.579619          1.280230        3.017075   
50%                              3.871914          1.771184        3.819156   
75%                              4.202333          2.124720        5.059843   
max                              6.312585          3.302401        6.962010   

       Health and Hospitals   Highways     Police  All Other  \
count             60.000000  60.000000  60.000000  60.000000   
mean               1.632759   1.243307   0.628770   5.103117   
std                0.915194   0.477030   0.150432   1.397332   
min                0.017351   0.010923   0.006636   0.049399   
25%                1.013766   0.968849   0.556597   4.304009   
50%                1.496192   1.096752   0.618976   4.935675   
75%                2.058508   1.505501   0.700228   5.447679   
max                4.788240   2.865624   1.116419  10.099438   

       Personal Income [1] (Millions)  All years income  \
count                    6.000000e+01      1.300000e+01   
mean                     9.250019e+05      1.433573e+07   
std                      2.482756e+06      2.489145e+06   
min                      3.450230e+04      1.060560e+07   
25%                      1.018024e+05      1.249270e+07   
50%                      3.042645e+05      1.399838e+07   
75%                      6.395509e+05      1.615188e+07   
max                      1.854226e+07      1.854226e+07   

       Police budget evolution  
count                13.000000  
mean              99998.421001  
std               13770.066801  
min               74133.109050  
25%               95017.083885  
50%               96875.949535  
75%              109116.131840  
max              123046.265209

Bien que les statistiques précédentes nous montrent une moyenne, une médiane et d'autres informations, nous les affichons tout de même ici sous une forme plus lisible.

In [81]:
print("===================Moyenne=====================")
print("All years income :", tax_sample["All years income"].mean())
print("Police budget :", tax_sample["Police budget evolution"].mean())

===================Moyenne=====================
All years income : 14335731.153846154
Police budget : 99998.42100129538
In [82]:
print("===================Médiane=====================")
print("All years income :", tax_sample["All years income"].median())
print("Police budget :", tax_sample["Police budget evolution"].median())

===================Médiane=====================
All years income : 13998383.0
Police budget : 96875.94953533
In [83]:
print("===================Skew=====================")
print("All years income :", tax_sample["All years income"].skew())
print("Police budget :", tax_sample["Police budget evolution"].skew())

===================Skew=====================
All years income : 0.3442130183886384
Police budget : -0.025606281986510004

Pour les graphiques nous avons choisi un graphique de courbe, car il est le plus explicite concernant les données sur une période en années. C'est ici que nous utilisons les listes contenant les valeurs récupérées lors de la boucle.

Évolution du revenu intérieur des USA¶

In [84]:
def formatter_1e6(x): # Pour l'affichage simplifié de l'axe Y
    return str(round(x / 1e6, 1))

tpc_total_fig, tpc_total_ax = pyplot.subplots()
tpc_total_ax.plot(numpy.array(budget_years), numpy.array(total_income_list))

# tpc_total_ax.yaxis.set_major_formatter(formatter_1e6)
tpc_total_ax.yaxis.set_minor_formatter(NullFormatter)
tpc_total_ax.set_xlabel("Année")
tpc_total_ax.set_ylabel("Budget")
tpc_total_ax.text(0, 1.05, "en millions ($)", transform=pyplot.gca().transAxes, ha="left", va="top")

tpc_total_ax.set_title("Évolution du revenu intérieur des USA")
pyplot.show()

Évolution du budget de la police aux USA¶

In [85]:
def formatter_1e4(x): # Pour l'affichage simplifié de l'axe Y
    return str(round(x / 1e4, 1))

tpc_police_fig, tpc_police_ax = pyplot.subplots()
tpc_police_ax.plot(numpy.array(budget_years), numpy.array(police_budget))
# tpc_police_ax.yaxis.set_major_formatter(formatter_1e4)
tpc_police_ax.yaxis.set_minor_formatter(NullFormatter)
tpc_police_ax.set_title("Évolution du budget de la police aux USA")
tpc_police_ax.set_xlabel("Année")
tpc_police_ax.set_ylabel("Budget ($)")
pyplot.show()

Analyse géographique des données¶

Carte¶

In [86]:
class Map:
	TILE_LAYERS = [
		{"name": "openstreetmap", "display_name": "Open Street Map"},
		{"name": "stamentoner", "display_name": "Stamen toner"},
		{"name": "cartodbpositron", "display_name": "CartoDB (Light)"},
		{"name": "cartodbdark_matter", "display_name": "CartoDB (Dark)"},
	]

	def __init__(self, crs="EPSG:3857", **kwargs):
		self.crs = crs
		self.map = folium.Map(tiles=None, crs=self.crs.replace(":", ""), **kwargs)
		self.layers = []

	def _add_tile_layers(self):
		for layer in Map.TILE_LAYERS:
			folium.TileLayer(layer["name"], name=layer["display_name"]).add_to(self.map)

	def _add_map_layers(self):
		for layer in self.layers:
			for sublayer in layer.get_layers():
				sublayer.add_to(self.map)

	def _register_layer(self, layer):
		if isinstance(layer, MapLayer):
			self.layers.append(layer)

	def fit_bounds(self, south_west, north_east):
		self.map.fit_bounds([south_west, north_east])

	def open(self, notebook=False, output_dir="./temp", filename=None):
		self._add_tile_layers()
		self._add_map_layers()
		folium.LayerControl().add_to(self.map)

		if notebook:
			return self.map

		path = os.path.join(output_dir, (filename if filename else f"map-{time.time()}.html"))

		self.map.save(path)
		webbrowser.open(path)

Couche de carte¶

In [87]:
class MapLayer:
	DATA_TYPES = {"DataFrame": 0, "Geo[GCS]": 1, "TimedGeo[GCS]": 1}

	def __init__(self, name, show_default=False):
		self.name = name
		self.parent_map = None
		self.feature_group = folium.FeatureGroup(self.name, overlay=True, show=show_default)
		self.layers = []

		self.data = None
		self.data_type = False

	def get_layers(self):
		return self.layers

	def add_to(self, m):
		if isinstance(m, Map):
			self.parent_map = m
			# noinspection PyProtectedMember
			self.parent_map._register_layer(self)

		return self

	def load_dataframe(self, data):
		self.data = data
		self.data_type = MapLayer.DATA_TYPES["DataFrame"]

		return self

	def load_gcs_data(self, data, col_names=None, time_column=None):
		if time_column is None:
			col_names = col_names if col_names else {"lat": "Latitude", "lon": "Longitude"}
			data_coords = data[[col_names["lat"], col_names["lon"]]].dropna(axis=0, how="any")
			geometry = geopandas.points_from_xy(data_coords[col_names["lon"]], data_coords[col_names["lat"]])

			self.data = geopandas.GeoDataFrame(data, geometry=geometry, crs=self.parent_map.crs)
			self.data_type = MapLayer.DATA_TYPES["Geo[GCS]"]
		else:
			data_coords = data[[time_column, col_names["lat"], col_names["lon"]]].dropna(axis=0, how="any")
			data_dates = data[time_column].drop_duplicates()
			data_timed = {}

			for _, d in data_dates.iteritems():
				data_timed[d.date().__str__()] = data_coords.loc[data_coords[time_column] == d][[col_names["lat"], col_names["lon"]]].values.tolist()

			self.data = data_timed
			self.data_type = MapLayer.DATA_TYPES["TimedGeo[GCS]"]

		return self

	def _add_to_layer(self, item):
		item.add_to(self.feature_group)
		self.layers.append(item)

	def to_heatmap(self, **kwargs):
		if not self.data_type == MapLayer.DATA_TYPES["Geo[GCS]"]:
			raise RuntimeError("MapLayer: to_heatmap() is only available for GCS data.")

		heat_data = [[point.xy[1][0], point.xy[0][0]] for point in self.data.geometry if not point.is_empty]
		heat_map = folium.plugins.HeatMap(heat_data, name=self.name, **kwargs)
		self._add_to_layer(heat_map)

		return self

	def to_timed_heatmap(self, **kwargs):
		if not self.data_type == MapLayer.DATA_TYPES["TimedGeo[GCS]"]:
			raise RuntimeError("MapLayer: to_timed_heatmap() is only available for timed GCS data.")

		heat_map = folium.plugins.HeatMapWithTime(list(self.data.values()), index=list(self.data.keys()), name=self.name, **kwargs)
		self._add_to_layer(heat_map)

	def to_choropleth(self, key_on=None, fill_color=None, **kwargs):
		if not self.data_type == MapLayer.DATA_TYPES["DataFrame"]:
			raise RuntimeError("MapLayer: to_choropleth() is only available for pandas dataframes.")

		choropleth = folium.Choropleth(
			data=self.data,
			key_on=(key_on if key_on else "feature.id"),
			fill_color=(fill_color if fill_color else "YlOrRd"),
			**kwargs
		)
		self._add_to_layer(choropleth)

		return self

Création de la carte¶

In [88]:
data_wp_geo = data_wp.loc[data_wp["is_geocoding_exact"]]
data_fe_geo = data_fe.loc[data_fe["Date of death"] >= datetime(2015, 1, 2)]
In [89]:
plot_map = Map()

encounter_layer = MapLayer("Fatal encounters", show_default=True)
encounter_layer.add_to(plot_map).load_gcs_data(data_fe, col_names={"lat": "Latitude", "lon": "Longitude"}, time_column="Date of death").to_timed_heatmap(
	min_opacity=0.4, radius=20
)

police_death_layer = MapLayer("Police Deaths", show_default=False)
police_death_layer.add_to(plot_map).load_dataframe(pd_state_data).to_choropleth(
	geo_data="https://raw.githubusercontent.com/python-visualization/folium/main/examples/data/us-states.json",
	columns=["State", "Number of deaths"],
	name="Police deaths",
	legend_name="Police deaths",
	key_on="feature.id",
	fill_color="YlOrRd",
	fill_opacity=0.7,
	line_opacity=0.2
)

plot_map.open(notebook=True)
Out[89]:
Make this Notebook Trusted to load map: File -> Trust Notebook